Abstract:Hyperspectral image (HSI) classification suffers from two serious problems, one is the limited labeled pixels, and the other is the class imbalance problem. As a result, the number of labeled pixels in many categories is not sufficient to characterize the spectral-spatial information, and train a satisfying deep model. By making full use of the information of unlabeled pixels, semi-supervised methods can provide better classification performance in the case of limited labeled pixels. However, they do not take into account the imbalance in the HSI data. As a method of data enhancement, generative adversarial networks focus on the above two problems and have also been widely used for the task of the HSI classification. In this work, we propose a distance constraints-based generative adversarial networks (DGAN) method for HSI classification to address these two problems. The DGAN employs the convolution autoencoder (AE) to extract the latent features of the HSI samples, and considers the reconstructed samples from the AE as the real samples for the later classifier and discriminator. In addition, the DGAN uses two distance constraints to solve the problems of the few labeled samples and class imbalance, the one latent-data distance constraint enforcing the generator to generate HSI samples for each class (especially the minority class), another discriminator-score distance constraint guiding the generator to synthesize samples that resemble the real HSI samples. Finally, the generated samples are combined classwise with the reconstructed samples and the real HSI samples to learn the parameters of the classifier and discriminator. Experimental results show that our method achieves state-of-the-art performance in terms of overall accuracy (OA) when trained with only 0.5%-4% of data sets from Indian Pines, Pavia University, and Botswana. Specifically, our method demonstrates improvements of 5.48%, 8.79%, and 0.91% on these three datasets, respectively. It reveals the gre- t potential of the DGAN model in generating the HSI samples for each class, which contributes to improving the classification performance of the HSI data.

An Optimized Training Method for GAN-Based Hyperspectral Image Classification

Distance Constraint-Based Generative Adversarial Networks for Hyperspectral Image Classification

Semisupervised hyperspectral image classification based on generative adversarial networks and spectral angle distance

Generative Adversarial Networks and Probabilistic Graph Models for Hyperspectral Image Classification

Generating images for Supervised Hyperspectral Image Classification with Generative Adversarial Nets

Hyperspectral Image Classification Based on Generative Adversarial Networks with Feature Fusing and Dynamic Neighborhood Voting Mechanism

Generative Adversarial Capsule Network With ConvLSTM for Hyperspectral Image Classification

Adaptive DropBlock-Enhanced Generative Adversarial Networks for Hyperspectral Image Classification

PGNN-Net: Parallel Graph Neural Networks for Hyperspectral Image Classification Using Multiple Spatial-Spectral Features

Self-Supervised Divide-and-Conquer Generative Adversarial Network for Classification of Hyperspectral Images

Graph Domain Adversarial Network With Dual-Weighted Pseudo-Label Loss for Hyperspectral Image Classification

Hyperspectral Image Classification Based on Superpixel Feature Subdivision and Adaptive Graph Structure

Cnn-assisted multi-hop graph attention network for hyperspectral image classification

Cross-Domain Hyperspectral Image Classification Based on Generative Adversarial Networks

Global and pyramid convolutional neural network with hybrid attention mechanism for hyperspectral image classification

Double-Branch Multilevel Skip Sparse Graph Attention Network for Hyperspectral Image Classification

Hypergraph Convolutional Network With Multiple Hyperedges Fusion for Hyperspectral Image Classification Under Limited Samples

Semisupervised Classification for Hyperspectral Images Using Graph Attention Networks

Physics-Based GAN With Iterative Refinement Unit for Hyperspectral and Multispectral Image Fusion

HDECGCN: A Heterogeneous Dual Enhanced Network Based on Hybrid CNNs Joint Multiscale Dynamic GCNs for Hyperspectral Image Classification

Improving GAN Training via Feature Space Shrinkage